Acute, synergistic and antagonistic effects of some aromatic compounds and essential oils on Planococcus citri by machine learning and feature selection approaches

Abstract

The individual and combined acute toxic effects of D-limonene, menthol, aromatic compounds, and essential oils from Salvia rosmarinus Spenn. , Mentha spicata L., and Salvia sclarea L. were evaluated against Planococcus citri (Risso) (Hemiptera: Pseudococcidae). Based on a comparison of lethal concentrations, menthol (LC 50 : 0.39 g/L) and M. spicata L. (LC 50 : 1.45 g/L) showed the highest efficacy, whereas D-limonene (LC 50 : 14.67 g/L) demonstrated the lowest. The volatile oil ingredients were identified using GC–MS/FID (Gas Chromatograph-Mass Spectrometer/Flame Ionization Detector). To predict acute and synergistic effects, the performance of several supervised machine learning algorithms was assessed using K-fold cross-validation. Among the models tested, Random Forest Regression yielded the best predictive performance. Feature selection algorithms indicated that D-limonene, menthol, alpha-pinene, carvone, 1,8-cineole, and sabinene were the primary contributors to acute toxicity. Analysis of the combination index (CI), t-distributed stochastic neighbor embedding (t-SNE), and heat maps revealed that mixtures of essential oils and aromatic compounds exhibited both synergistic and antagonistic interactions. Notably, structurally similar molecules such as limonene and its cis- and trans-isomers, α-pinene and camphene, camphor and 1,8-cineole, and borneol and isoborneol were predicted to exert predominantly synergistic effects. Machine learning and traditional methods reveal that the synergistic effects of complex essential oil mixtures on P. citri remains insufficiently understood. Aromatic compounds and essential oils show diverse biological activities against P. citri and may serve as viable alternatives to synthetic pesticides in pest management strategies. Furthermore, machine learning and deep learning may offer more rapid predictions and insights into toxicity profiles.